Should AI Companies Be Supervised Like Banks?
Ask about Knowledge at Wharton
Episode Description
ABOUT THE EPISODE AI has already permeated nearly every part of society and the economy, but how should we manage the risks as the technology becomes more powerful?
Peter Conti-Brown, Professor at the Wharton School, explores whether bank supervision could offer a useful model for overseeing AI companies.
The conversation examines the difference between regulation and supervision, the possibility of AI stress tests, and how regulators could prepare for scenarios ranging from model failures to geopolitical attacks.
But banking isn't a perfect blueprint. AI presents unique risks, raising a fundamental question: when private companies take enormous risks developing AI, who ultimately holds the bag?
ABOUT THE PODCAST This Week in Business features interviews with Wharton faculty about the latest news, fascinating trends, and issues impacting both consumers and the business world. Episodes are recorded at the Wharton School and published twice per week.
----- The Wharton School of the University of Pennsylvania is committed to sharing its intellectual capital through Knowledge at Wharton, the school’s online business analysis journal.
Launched in 1999, Knowledge at Wharton offers free access to:
- Articles, podcasts, and videos highlighting Wharton faculty research and analysis of current business trends - Interviews with book authors from Wharton School Press - In-depth series of curated content like Wharton Executive Education’s Nano Tools for Leaders - A searchable database of more than 10,000 articles covering all aspects of business - A weekly newsletter that delivers Knowledge at Wharton insights directly to your inbox
Visit the Knowledge at Wharton homepage: https://whr.tn/3qUpSBE Subscribe to the Knowledge at Wharton YouTube channel: https://whr.tn/3DEAgAU
Read the full transcriptHide the full transcript
- Well, certainly we know there are still many questions to be asked and answered in and
- around the world of artificial intelligence. One big one is, of course, what the regulatory
- landscape is going to look like. There are many potential paths that we may see play out in that
- area. One man to ask about it, ironically, was involved in some research that looked into this
- earlier this year, and that is Peter Conti-Brown, professor at the Wharton School,
- and he joins us to talk to us about why banking regulation may play a role in this process.
- Peter, great to have you with us today. Thanks very much for your time. Delighted to be here with you, Dan.
- I find this very interesting because you and I have talked so much about the role of banking
- regulation, the role of the Fed, but in this scope, I guess it is an instance where it's
- an open canvas at the moment, and we're looking for the best potential way to go.
- And you and our friend Kevin Werbach talked about why the framework of banking regulation
- may be a good path to follow. That's right. And to be clear, we were thinking more bank supervision, a bigger distinction,
- regulation, which exists in banking and in food safety and in AI and all kinds of things. Those
- are the rules that Congress either makes or creates agencies to make that govern everybody.
- Supervision is about trying to create this apparatus where we have public officials and
- private actors engaged in ongoing conversations about how to manage risk. Bank supervision's
- been around for a really long time and since the Civil War. And we thought that there were
- some applications from that field to, you know, this crazy wild west world of AI that might be
- more useful than trying to write rules once and pretend that they're going to be useful to us,
- you know, five years later, a year later, a month later. What specifically might carry over from bank supervision
- to what we may see play out with artificial intelligence here? I think it's important to think about the fact that with banking, what we have is a big
- asymmetry. We have banks that really want to take risks that benefit their shareholders,
- benefit their own stakeholders, but also grow the economy. A profitable bank, mostly, not always,
- mostly, is doing things that are going to, is going to, that will profit their community as
- well. But there's a big downside. So when they become, take these big bets that go south,
- who ends up holding the bag? Well, you and I have talked many times over the last decade about,
- about the, you know, Joe Q public, the bag holder of that risk, the bailouts. And AI is pretty similar. So we have AI companies that are just
- pushing on the frontier. We even use that word, the frontier models of AI, wanting to do more and greater and bigger things, taking risks that can
- show what the, this technology might do. And the problem is, is that what happens on the downside,
- who holds the risk then? The idea is that bank supervision or AI supervision is about trying to
- manage that backend risk, that residual risk. Try to understand that when the private sector is
- taking risks to grow the economy, grow these models, push that frontier, they're doing so in
- a way that reflects the best interests, not just of their stakeholders, but of everybody.
- Which is interesting because I think there have been numerous instances when you look at
- how things play out, it ends up being the consumer that ends up holding the bag. And that is the
- last thing you want to see occur here, especially in something that by so many people's account is
- going to be transformative for our culture, but also has such a huge investment in it as well.
- Yeah. I mean, I think that's really it. It's just something that's fun for me to do when I'm
- teaching like executives or even my own MBAs who are much younger, is really just asking the
- question, you know, let's take two technological innovations in the past few years about which
- we've spent a lot of time talking, crypto and AI. And my question to them is, when was the last time
- you used a cryptocurrency to settle a transaction for which you'd otherwise use a US dollar? In
- other words, not to invest in crypto as an asset and not to buy more crypto, but because you wanted
- to buy a stick of gum or a pair of shoes or a car or whatever you needed to use crypto.
- The number of people who raised their hands is extremely small, not zero, but almost zero.
- And then I'll ask students, when was the last time you used AI? How many of you used AI in the last
- month? Every single hand goes up. Last week, every hand goes up. Last day was the last time I asked
- this question. Every hand went up. And so what we've got now is this AI revolution that has
- permeated just about every corner of our society and economy. And the question is, how are we
- going to manage those risks when things are great, but also when things are not? And that's what
- kind of this idea of using the model of supervision from banking is meant to do. And the banks are
- privately owned, and we assume that all of the AI companies are going to be ones that are going to
- be at worst privately owned, assuming that, you know, until they go to a public market.
- But there are links from that perspective as well, correct? Oh, yeah. I mean, the ownership structure of AI companies is a little bit diversified.
- Mostly they're privately held companies, but some of them have non-profit boards or a non-profit,
- like OpenAI has a non-profit organization that is affiliated with the for-profit company.
- Some of them have pledged, like Anthropic has pledged to have kind of a super board of
- public officials meant to manage those risks. But, you know, we have banks like that too. Banks
- come in a huge number of varieties of corporate structures from member-owned credit unions to
- benefit corporations that are meant to serve some sort of purpose other than profitability for shareholders. But by and large, banks are publicly held
- corporations. And the idea is that supervision can be neutral on the type of corporate governance
- that an institution might have. And it's better to orient that supervision toward the kinds of risks
- that the institutions are taking. You noted in the article that you and Peter wrote, or I should
- say Kevin wrote a few months ago, that the structure ends up being bank examiners kind of
- sitting within the company, within the bank, to be able to see what's going on. And I guess the
- question I have is, are the AI companies, the way they are structured, willing and able to kind of
- go down that path as well? No, willing and able are different things, right? Yeah. Able for sure. I mean, the way that the companies exist, mostly it's just code. It's just
- giving access to understanding how the frontier models work and sharing that access,
- including subject to strict rules of confidentiality with public officials who have the expertise relevant to understand what it is they're looking at.
- Those are non-trivial problems. You don't want to have private corporations who have
- built these extraordinary things worried that they can't trust their supervisors,
- they can't trust the government with these kinds of secrets. But we're kind of past that point
- just through the contract relationship as counterparties with the defense department,
- for example. That's been a big example where the government is already in a position of
- supervising some of these risks. And the question is, can we structure that supervision in a way
- that's a little bit more coherent, a little bit more effective? You say one of the things that
- AI companies are going to have to do is make sure that they are strong and steady moving forward.
- So a version of basically the stress tests that we have seen become very important in the banking
- system is something that would probably do very well by the AI companies moving forward.
- Yeah, I think that's right. What do you think is the value of having that when you're talking about an AI company?
- Well, the main benefit of a kind of stress test sort of ecosystem is where you create a scenario
- analysis and say this is not only how have you performed to date, how do you expect to perform
- in the future, but how do you expect to perform in the future subject to these kinds of parameters?
- So in the banking context, although the concept of a stress test has evolved a lot over the last
- 15 years that we've been doing it, has a lot that allows us to kind of play with counterfactuals
- and use our imagination to think, how is this going to go? And that's kind of where we are
- with AI is thinking, where is this going to go? And what would happen if we did this, we did that?
- What would happen if the AI burst through these kinds of settings? What would happen if it was
- hijacked by a geopolitical adversary? There's kinds of questions you can ask and then you
- can play out and war game those scenarios. Well, and to that degree, obviously some of
- the conversation that's going on right now is about the AI models that would be coming from
- China and the uncertainty we have on how those are operated and how they would fit into the
- structure of, you know, people using AI models here, which is a little bit different than the
- banking structure, isn't it? Oh yeah, exactly. I mean, that's a, it's an important point that
- you're putting your finger on, which is this model doesn't work perfectly. We're not saying that
- Kevin and I are not saying, Hey, look, AI is hard. The risks are hard, but you know, we solved
- this problem once and for all and banking, we should just do the same thing over again, because
- number one, we didn't solve anything in banking. We still have major problems there. Right. And so
- it's not necessarily a model that you're going to apply wholesale without, without thought.
- And the second is, you know, some of these risks really are different. They're unique.
- So they'd require some, some real specialization and real, real thoughtfulness as you, as you move through it. Peter, great to talk to you today. Thanks
- very much for your time. All the best. Always a pleasure, Dan. Thank you. Thank you. Peter Conti-Brown of the Wharton School.
Brief Summary
Peter Conti-Brown discusses AI regulation and banking supervision's role in managing risks.
Key points
- The Role of Banking Regulation in AI. Peter Conti-Brown discusses how banking regulation could inform AI supervision.
- Consumer Risks in AI. Peter highlights the potential risks consumers face as AI evolves.
- AI Usage in Daily Life. Peter engages students on their experiences with AI versus cryptocurrency.
- The AI Revolution. Peter emphasizes the pervasive impact of AI across society and economy.
- Unique Risks of AI. Peter discusses the unique challenges in regulating AI compared to banking.
Episode highlights
- The Role of Banking Regulation in AIPeter Conti-Brown discusses how banking regulation could inform AI supervision.“AI is a crazy wild west world.”1:50Aug 19, 2026
- Consumer Risks in AIPeter highlights the potential risks consumers face as AI evolves.“The consumer ends up holding the bag.”3:35Aug 19, 2026
- AI Usage in Daily LifePeter engages students on their experiences with AI versus cryptocurrency.“When was the last time you used AI?”4:44Aug 19, 2026
- The AI RevolutionPeter emphasizes the pervasive impact of AI across society and economy.“AI revolution has permeated every corner of our society.”4:58Aug 19, 2026
- Unique Risks of AIPeter discusses the unique challenges in regulating AI compared to banking.“Some risks really are different. They're unique.”10:50Aug 19, 2026





